The gut microbial metabolite formate exacerbates colorectal cancer progression

Abstract

The gut microbiome is a key player in the immunomodulatory and protumorigenic microenvironment during colorectal cancer (CRC), as different gut-derived bacteria can induce tumour growth. However, the crosstalk between the gut microbiome and the host in relation to tumour cell metabolism remains largely unexplored. Here we show that formate, a metabolite produced by the CRC-associated bacterium Fusobacterium nucleatum, promotes CRC development. We describe molecular signatures linking CRC phenotypes with Fusobacterium abundance. Cocultures of F. nucleatum with patient-derived CRC cells display protumorigenic effects, along with a metabolic shift towards increased formate secretion and cancer glutamine metabolism. We further show that microbiome-derived formate drives CRC tumour invasion by triggering AhR signalling, while increasing cancer stemness. Finally, F. nucleatum or formate treatment in mice leads to increased tumour incidence or size, and Th17 cell expansion, which can favour proinflammatory profiles. Moving beyond observational studies, we identify formate as a gut-derived oncometabolite that is relevant for CRC progression.

Fig. 1. Fusobacterium levels are elevated in stool and tissue from patients with CRC and are associated with CMS1 and metabolic-driven CMS3.

a, Differential abundance of bacterial genera in CRC stool samples. Clustered heatmap (Pearson correlation) shows the significant differentially abundant bacterial genera in an in-house cohort of samples from patients with CRC (n = 52 independent biological samples) in comparison with healthy donor (n = 63 independent biological samples) (P < 0.05; negative binomial Wald Benjamini–Hochberg testing) as identified by 16S rRNA gene sequencing. Top CRC-enriched bacteria are marked in red and separately plotted as abundance (baseMean) of a bacterium per patient (right). Heatmap intensities (blue colour scale) represent log₁₀ base mean values. b, Differential abundance of bacterial genera in CRC-tissue samples, following a PathSeq analysis on whole-exome sequencing (WXS) data (TCGA). Clustered heatmap (Pearson correlation) showing the significant differentially abundant bacterial genera in tumour tissue samples from patients with CRC (n = 50) in comparison with adjacent healthy mucosal samples (n = 50, P < 0.05; negative binomial Wald Benjamini–Hochberg testing). Top CRC-enriched bacteria are marked in red and separately plotted as abundance (score) of a bacterium per sample (right). Heatmap intensities (blue colour scale) represent log₁₀ scores. c, Fusobacterium abundance in tissue samples from patients with CRC (n = 595 independent biological samples, TCGA). WXS data were analysed for the logarithmic score abundance of Fusobacterium across all samples via quantile-based classification (colour code). d, Fusobacterium abundance across CMS subtypes. The cohort (c) was subjected to gene expression analysis and CMS classification of matching RNA-seq. Coloured, segregated bars show the proportion of patients with fusobacterial loads per CMS. Significant differences were observed for CMS1 versus CMS2 (P = 0.003103733) or CMS4 (P = 0.006062911) and CMS3 versus CMS2 (P = 0.038151361) or CMS4 (P = 0.035902419) Chi-squared tests. n^CMS1 = 62, n^CMS2 = 207, n^CMS3 = 69, n^CMS4 = 139 biologically independent samples. Source data

Fig. 2. Gut-on-chip cocultures (HuMiX) of Fn with patient-derived primary CRC cells reveal protumorigenic and pro-invasive effects along with an altered metabolic profile.

a, Schematic representation of the HuMiX setup. b, IPA canonical pathway analysis of HuMiX HT-29 co- versus monoculture differential gene expression analysis. Plot shows z-scores, -log(P) values and the number of molecules per pathway. Selected significant pathways are shown (P < 0.05), Fisher’s Exact Test. c, IPA Diseases or Functions analysis of HuMiX HT-29 co- versus monoculture differential gene expression analysis. Plot shows z-scores, P values, and the number of genes per pathway. Top ten significant pathways are shown (P < 0.05): Fisher’s exact test. d, Exometabolite profile of HuMiX cocultures. Untargeted GC–MS metabolite detection in chamber supernatants was used to measure relative, normalized metabolite levels. Heatmap shows relative log intensities of extracellular metabolites (rows) accross n = three independent experiments, each with one device per condition (main columns) composed of four chambers (subcolumns). e, SCFA levels in the HuMiX after HT-29-Fn coculture. Heatmap shows mean concentrations of extracellular SCFAs (rows) from n = 3 independent experiments, each with one HuMiX device per condition, composed of four chambers (columns). f, Levels of formate in the HuMiX chambers using different Fn isolates (Fn 23726 and one clinical isolate Fn ssp. animalis 7_1) in coculture with HT-29 or Caco-2 cells (n = 1 experiment with one HuMiX device per condition). ND, not detected and n.s., not sampled. Source data

Fig. 3. In silico modelling delineates enhanced formate metabolism of Fn in CRC context, which is also observed in Fusobacterium-high patients with CRC.

a, Metabolic models of the central carbon metabolism of Fn (left) and HT-29 cells (right) in coculture versus monocultures. Biomass production was 0.20476 for Fn and 0.01 for HT-29. Flux constraints were set according to the metabolite secretion profiles of the HuMiX monoculture controls and pairwise interactions were calculated via FVA. Heatmap arrows show maximum flux changes (mmol dGW⁻¹ h⁻¹) of depicted reactions as predicted by FVA. Arrows at the cells’ boundaries represent intra-extracellular exchange reactions. Measured differentially abundant exometabolites in the respective HuMiX coculture chambers are shown as filled circles (blue, decreased; red, increased). Extracellular filled circles show unassigned metabolites. Coloured arrow borders show upregulated (red) or downregulated (blue) genes involved in the respective reaction fluxes, on the basis of RNA-seq data (refer to Fig. 2b). Dashed arrows show indirect connections between nodes. 2HB, 2-hydroxybutyrate; AC, acetate; ACAC, acetoacetate; ATMP, adenylthiomethyl-pentose; BUTOH, butanol; EAAT, excitatory amino acid transporters; GLYAC, glyceric acid; LAC, lactate; MCT, monocarboxylate transporters; NAcP, n-acetylputrescine; PAGN, phenylacetylglutamine; PGA, pyruglutamic acid; and bdhAB, butanol dehydrogenase. b, Core set of Fn-related metabolites. Venn diagram shows overlapping, differentially abundant metabolites in the Yachida dataset (purple), in HuMiX (yellow) and in the metabolic model (green). c, IPA network analysis of the gene-regulatory role of Fn core metabolites (red) in connection with host gene-regulatory nodes of stemness and invasion. d, KEGG orthology gene abundances of Fn-related genes identified by the model in patients with CRC with different fusobacterial load in stool samples. Two-tailed Spearman’s rho correlation testing was used. e, pfl in stool metagenomes of stage I/ II, Fn^high patients with CRC. n^Fn-no = 19, n^Fn-low = 34, n^Fn-high = 58 (NS, not significant, P = 0.0219 for Fn-no versus Fn-high; one-way analysis of variance (ANOVA) with Tukey’s honestly significant differences test). *P < 0.05. Source data

Fig. 4. Formate drives CRC cell invasion by increasing cancer stemness.

a, Scratch wound healing capacity of HCT116 CRC cells over 48 h, n = 3 biologically independent experiments. Gut formate physiological dose is considered to be near 10mM. b, Transwell invasion of HCT116 CRC cells at 48 h. Data show the means of replicates from n = 5 independent experiments, P = 0.0079, two-tailed Mann–Whitney test. c, Focal adhesion formation in HCT116 CRC cells after formate or rhWnt3A exposure for 24 h, Data show means of technical replicates with at least 100 cells per condition from two independent experiments (indicated by different shapes). P = 0.0150, P = 0.0037 and P = 0.0067 for formate 1 mM, formate 10 mM and rhWnt3A versus control, respectively, ordinary two-way ANOVA with Tukey’s multiple comparisons test. d, ALDH1A1 protein levels in formate (+) treated versus untreated (−) HT-29 cells. e,f, ALDH activity in human (e) or APCmin mouse colonic organoids (f) after formate stimulation (5 mM, 24 h; FACS). Left, representative histogram of the ALDH⁺ cell population (crossbar). Right, quantification of ALDH activity, n = 3 independent experiments. P = 0.0053 in e and P = 0.0129 in f, unpaired two-tailed t-test. FITC, fluorescein isothiocyanate. g, Schematic representation of experimental setup. Derived control or formate-treated organoids from a CRC patient were dissociated and reseeded at 24 h. h, The organoid formation capacity on day 10 after reseeding. Left, the number of organoids counted in one representative experiment out of two, with three technical replicates, Right, representative images of one well per condition, i, Normalized organoid formation, n = 2 biologically independent experiments (different data point shapes) with two or three technical replicates per condition. j, Schematic overview of the intravenous metastatic dissemination model. HT-29-Luc tumour cells were treated with formate before intravenous injection into NSG mice (1 × 10⁶ cells per 200 µl per injection). k, IVIS imaging after luciferin injection on day 30 after tumour cell injection. Left, representative images from one mouse per group. Middle, the quantified tumour cell signal from lungs. Data show the total luciferase signal (reported as photon flux per mg of tissue), normalized to control, n = 6 biologically independent animals per condition, pooled from two independent experiments. Right, the number of lung macroscopic metastatic nodes, n = 3 biologically independent animals per condition from one representative experiment out of two. P = 0.0257 and P = 0.0045 in the middle and right panels, respectively, unpaired two-tailed t-test. b, c, e, f and k are shown as mean ± s.e.m. *P < 0.05, **P < 0.01. Source data

Fig. 5. Microbiome-derived formate drives metastatic dissemination through the activation of the AhR signalling pathway.

a, IPA analysis of differentially expressed genes in formate (10 mM) versus PBS treated HT-29 cells. Plot shows z-scores, P values and the number of molecules per selected significant pathway (−log(_P) > 1.3), Fisher’s Exact Test. b, AhR nuclear translocation in HT-29 cells treated with FICZ (known AhR ligand, 100 nM) alone or combination with formate (10 mM) for 6 h. Data shows technical replicates from two independent experiments, P = 0.00113, two-tailed nested ANOVA. Dashed lines represent the medians and dotted lines represent the upper and lower quartiles. c, Transwell invasion of HCT116 CRC cells after formate stimulation (10 mM, 48 h) alone, or in presence of an AhR signalling inhibitor (CH223191, 0.5 µM); P = 0.0417 for formate versus control; ordinary one-way ANOVA. d, Transwell invasion of HCT116 CRC cells after Fn and Escherichia coli coculture (MOI 10, 2 h). ***P < 0.001; ordinary one-way ANOVA. e, Transwell invasion of HCT116 cells upon Fn preinfection alone or in presence of an AhR signalling inhibitor (CH223191, 0.5 µM) for 24 h: paired two-sided t-test. c–e, Data show pooled means of replicates from three (d and e) or four (c) independent experiments. f, Schematic representation of experimental setup of g and h: NSG mice were intravenously injected with 1 × 10⁶ cells per 200 µl (g) or 0.5 × 10⁶ cells per 200 µl (h) HT-29-Luc cells, preinfected with Fn (MOI 10) for 2 h and pretreated with (h) or without (g) AhR inhibitor for 24 h (CH223191, 0.5 µM). After 30 d, signals of HT-29-Luc cells were determined using the IVIS. g,h, Left, representative image of one mouse per group. Middle, data show the total luciferase signal (reported as photon flux per mg of tissue) normalized to control. Right, the number of lung macroscopic metastatic nodes, n = 6 biologically independent animals per condition in g and h, excluding (g) (middle), where n = 5 biologically independent animals per condition (one mouse failed luciferin intraperitoneal injection). Regions of interest in g: control, 1–11.001 × 10⁶ and Fn, 2–3.379 × 10⁶. P = 0.0094, P = 0.0110 and P = 0.0210 in g (middle and right) and h (right), respectively, unpaired two-tailed t-test. Total flux was measured in the chest area. Data in c–e, g and h are represented as mean ± s.e.m. *P < 0.05, **P < 0.01, ***P < 0.001. Source data

Fig. 6. Fn and formate increase cancer stemness in vivo.

a, Schematic overview of the tumour infection mouse model. SPF NSG mice were subcutaneously injected with HT-29 cells subcutaneously (1 × 10⁶ cells per flank). After tumour formation, tumours were injected with Fn (MOI 10) or PBS (ctrl) or formate (10 mM, 60 μl) for five consecutive injections over 11 d. b, Formate levels in TIF of Fn-injected tumours. TIF volume was only sufficient for processing from n = 4 independently treated tumours per group. c, Gene expression levels of AHR (far left), ALDH1A1 (left), CYP1B1 (right) and SOX2 (far right) as assessed by rt–qPCR for n = 5 independently treated tumours per group. d, ALDH activity assay of Fn-/PBS-infected tumours. Left, representative histogram of ALDH⁺ cell populations (crossbar) in mouse tumours assessed by FACS. Right, quantification of ALDH activity in tumours on Fn infection. e, CD113 (left) and CD24 (right) expression in mouse tumours after intratumoral injection, as assessed by FACS, n = 8 independently treated tumours per group in d and e. f, Serial transplantation of Fn-treated xenografts. Tumours were explanted at endpoint, dissociated in culture and reinjected subcutaneously in secondary recipient mice at 5,000 cells per flank, n = 6 biologically independent animals per group. Kaplan–Meier analysis of tumour incidence was performed using a Mantel–Cox test, P = 0.0024. g, Expression of ALDH (left), CD44 (middle) and OCT4 (right) in formate-treated and control xenografts as assessed by FACS, n = 10 and n = 12 independently treated tumours for the control and treated conditions, respectively. h, Untargeted metabolite analysis of TIF samples from Fn-/PBS-infected xenografts. Heatmap shows normalized intensities, n = 8 and n = 6 independently treated tumours for the control and treated conditions respectively. Data in b–d (right), e and g are shown as mean ± s.e.m, P = 0.0006 in b, P = 0.0051 in c (far left), P = 0.0482 in c (right), P = 0.0262 in c (far right), ***P < 0.001 in d (right), P = 0.0052 in e (left), P = 0.0380 in e (right), P = 0.0209 in g (left), P = 0.0409 in g (middle) and P = 0.0450 in g (right), unpaired two-sided t-test. *P < 0.05, **P < 0.01, ***P < 0.001. Source data

Fig. 7. Fn administration and formate treatment lead to an increase in tumour incidence or tumour size and an increase in Th17 cells.

a, Schematic overview of the experimental setup. Germ-free mice received a single dose of AOM intraperitoneal (10 mg kg⁻¹). After 3 days, mice were gavaged with bacteria (10⁸ CFU per mouse) and euthanized at 8 weeks postinjection. b, Colonic tumour incidence, n = 9 and n = 10 biologically independent animals in the control and treated groups respectively, pooled from two independent experiments. c, Immune cell phenotyping of Fn versus PBS gavaged mice. Heatmap shows estimate (log₂ fold change (FC)) of normalized immune cell counts in mouse lamina propria (LP), MLNs and spleens. Filled star indicates P < 0.05, two-tailed least squared means method, d, Th17 cell counts (CD4⁺IL-17⁺RORγT⁺ T cells) in LP from c. One representative experiment is shown with n = 7 and n = 5 biologically independent animals in the control and treated groups respectively, in c and d. e, OCT4⁺ cells per colonic crypt, n = 3 biologically independent mice, with at least eight crypts per mouse, f, Schematic overview of the experimental setup. Germ-free (GF) mice received a single dose of AOM intraperitoneal (10 mg kg⁻¹) and formate was administered via the drinking water (250 mM) for the duration of the experiment (14 weeks). g,h, Tumour incidence (g) and surface areas (h) in colons of formate-treated mice and controls. i, Immune cell phenotyping of formate-treated mice and controls. Heatmap shows estimate (log₂ fold change) of normalized immune cell counts in the MLNs or spleens. Filled star indicates P < 0.05, two-tailed least squared means method. j, Th17 counts (CD4⁺IL^-17⁺RORγT⁺) in the MLNs from i. n = 11 biologically independent animals per group from two independent experiments pooled for representation in g,i and j. h, One representative experiment with all tumour surface areas from four independent biological animals per condition. The solid line represents the median and the dotted line represents the upper and lower quartiles. Data in b,d,e,g and j are shown as mean ± s.e.m. P = 0.0004 in b, P = 0.0095 in d, ***P < 0.0001 in e, P = 0.0363 in j, unpaired two-tailed t-test, P = 0.00259 in h, two-tailed nested ANOVA (factor mouse nested within condition). *P < 0.05, **P < 0.01, ***P < 0.001. Source data

Fig. 8. Microbiome-derived formate drives CRC invasion via secreted formate.

The Fn metabolism product formate promotes AhR signalling in vitro and in vivo, which results in increased Th17 cell infiltration and ALDH activity, cancer stemness and metastatic dissemination of tumour cells.

Extended Data Fig. 1. Microbiome composition in CRC patients and pathways identified in Fusobacterium^high patients.

a, Abundance of top CRC-enriched and differentially abundant bacteria in tissue per stage (TCGA). Boxplots shows medians with 1^st and 3^rd quantiles. The whiskers from the hinges to the smallest/largest values represent 1.5*inter-quartile range (IQR). n=108 healthy, n=102 Stage I, n= 209 Stage II, n=163 Stage III and n=86 Stage IV biologically independent donor samples, p=0.000705, p=0.00101, p=0.0000154 and p=0.00153 for Bacteroides in Stages I-IV vs. Healthy respectively, p=0.0112 and p=0.0402 for Campylobacter in Stages II and IV vs. Healthy respectively, p=0.0223 for Fusobacterium in Stage II vs Healthy, p=0.0293, p=0.000608 and p=0.00267 for Gemella in Stages I, II and IV vs. Healthy respectively, pairwise t test. b, Fn abundance in the EGA cohort. PathSeq-analysis for Fn abundance in matching normal vs. adenocarcinoma tissue on RNA-seq data, p=0.00165, paired two-tailed t test. c, Fn tissue abundance distribution in CRC patients of the EGA cohort. Fn was detected via RNA-seq and analyzed for the logarithmic score bacterial distribution (n=69). Quantile-based classification (color code) was applied for the target bacterium (quantcut function from the gtools R package). d, Correlation of Fn abundance with consensus molecular subtypes in CRC. The cohort in c was subjected to gene expression analysis and further classified via the CMScaller R package into CMS as described. Colored, segregated bars show the proportion of patients with differing fusobacterial loads per CMS. Chi-squared tests were performed for comparing the fusobacterial across all CMS. No significant differences were observed. n^CMS1=17, n^CMS2=19, n^CMS3=7, n^CMS4=5. e, IPA analysis of Fusobacterium^{high vs. no} differential gene expression analysis of the TCGA dataset. Plot shows z-scores, p-values, and the number of molecules per pathway. Selected significant pathways are shown (-log(p-value)>1.3). f, KEGG-based GSEA of Fusobacterium^{high vs. no} differential gene expression analysis of the TCGA dataset (pathfindR R package). Plot shows fold enrichment, p-values, and the number of genes per pathway. All significant pathways are shown (p<0.05). g, IPA analysis of Fusobacterium^{high vs. no} differential gene expression analysis of the EGA dataset. Plot shows z-scores, p-values, and the number of enriched molecules per pathway. Selected significant pathways are shown (-log(p-value)>1.3). *p<0.05, **p<0.01, ***p<0.001. Source data

Extended Data Fig. 2. Microbiome cross-talk studies in HuMiX.

a, Experiment layout of HuMiX experiments. Devices assembly and medium priming for 1 day; epithelial cell inoculation and monolayer formation for 6 days; bacterial-human coculture for 24h. Control devices with either bacterial or human cell mono-cultures were run in parallel. At the endpoint, devices were opened for chamber metabolite sampling and cell harvesting. b, Clamped HuMiX device, showing the HuMiX device (spiral), the N₂ and medium lines as well as free optode pockets (cf. Figure 2A). c, Oxygen levels (%) as monitored in the perfusion and N₂ gas microchambers of HuMiX HT-29 mono-/ and cocultures. The grey box marks the the co-culture time. d, Fn and HT-29 cell viability in HuMiX mono-/ and co-cultures. Human cell viability was assessed by TrypanBlue staining and FACS-based L/D NearIR staining. Bacterial cell viability was assessed by BacLight L/D staining under a microscope. Three representative images were taken and cell viability was assessed with ImageJ. (c), (d) and (e) show data from n=three biologically independent experiments. e, Cell counts of human cells and bacteria under co-culture. Cell counts of HT-29 cells were taken after TrypanBlue staining. Bacteria cell counts were assessed using CASY cell counter with the 45µm capillary. f, Western Blot detection of MAPK signaling, as well as NF-κB signaling on HuMiX HT-29 cell lysates. phosphorylation of MEK at Ser217/221, p65 at Ser536, pERK at thr 202/tyr204, and p38 at thr180/tyr182 under HT-29 monoculture (-) and HT-29-Fn co-culture (+) conditions in HT-29 cells. Data shows one representative of three biologically independent experiments. g, Quantified ratios of phosphorylated protein per total target protein levels from (f), n=three biologically independent experiments. All boxplots show medians with 1^st and 3^rd quantiles. The whiskers from the hinges to the smallest/largest values represent 1.5*inter-quartile range (IQR). ns=not significant; two-sided paired t-tests. h-i, Correlation of global transcriptomic profile of HuMiX co-cultures with Fusobacterium^{high vs. no} patient’s gene expression profile of the TCGA (h) and EGA (i) datasets. Interesting gene targets are marked in red. Two-tailed Spearman’s rho correlation testing was used to assess significance between the two data sets. Source data

Extended Data Fig. 3. Metabolome composition in HuMiX microbiome-host crosstalk.

a, Significant differentially abundant metabolites in HT-29 mono-/ vs. HT-29-co-cultures, human cell microchamber supernatants. b, Significant differentially abundant metabolites in (1) HT-29-co-cultures comparing inflow metabolite levels with human chamber metabolite levels, (2) Fn-co-cultures, comparing inflow metabolite levels with bacterial microchamber supernatants. (3), HT-29-co-cultures and Fn-co-cultures, comparing inflow metabolite levels with human and bacterial microchamber supernatants. These were ambiguous metabolites that could not be assigned to any of the two cell types as being produced or secreted. c, Significant differentially abundant metabolites in Fn mono-/ vs. Fn-co-cultures, bacteria microchamber supernatants. a, b and c show data from n=three biologically independent experiments. All boxplots (in a, b and c) show normalized median metabolite levels (log intensities). Boxplots show medians with 1^st and 3^rd quantiles. The whiskers from the hinges to the smallest/largest values represent 1.5*inter-quartile range (IQR). p<0.05 for a, b, and c; unpaired, two-sided t-test. d, Levels of formate in different Fn isolates (Fusobacterium nucleatum 23726 and Fusobacterium nucleatum ssp. animalis 7_1, clinical isolate, one chamber per isolate) in monoculture and co-cultures with T 18 cells in HuMiX. e, Levels of metabolites in HuMiX in HT-29 cells alone or in co-culture with different Fn isolates (Fusobacterium nucleatum 23726 and Fusobacterium nucleatum ssp. animalis 7_1, clinical isolate, one chamber per isolate). f, Genes related to AhR signaling and cancer stemness in HT-29 cells after co-culture with different Fn isolates in HuMiX. Source data

Extended Data Fig. 4. In silico metabolic modelling of Fn-host tumor cell interaction.

a, Maximum flux changes for exchange reactions of F. nucleatum secreted metabolites. Calculated flux ranges from the FVA results of the F. nucleatum monoculture model were subtracted from FVA results of the HT-29-Fn co-culture model. No minimum flux changes were observed. b, Minimum and maximum flux changes for exchange reactions of HT-29 secreted/uptaken non-essential amino acids and other metabolites. Calculated flux ranges from the FVA results of the HT-29 mono-culture model were subtracted from flux ranges of the HT-29-Fn co-culture model. c, Minimum and maximum flux changes for exchange reactions secreted/uptaken essential amino acids in the HT-29 model. Calculated flux ranges from the FVA results of the HT-29 mono-culture model were subtracted from flux ranges of the HT-29-Fn co-culture model. d, Minimum and maximum flux changes for intracellular reactions of F. nucleatum. Calculated flux ranges from the FVA results of the F. nucleatum mono-culture model were subtracted from FVA results of the HT-29-Fn co-culture model. Cutoff for displayed reactions was -2 and 2 mmol*dGW^-1*hr^-1. e, Minimum and maximum flux changes for intracellular reactions of HT-29. Calculated flux ranges from the FVA results of the HT-29 mono-culture model were subtracted from flux ranges of the HT-29-Fn co-culture model. Cutoff for displayed reactions was -10 and 10 mmol*dGW^-1*hr^-1. Source data

Extended Data Fig. 5. Fusobacterium nucleatum and formate induced pathways.

a, Focal adhesion formation in HT-29 cells after formate or rhWnt3A (100ng/mL) exposure for two days. Data shows technical replicates with at least 100 cells per condition from three independent experiments (different shapes), p=0.0029240 and p=0.0024919 for formate 10mM and Wnt3A vs. control respectively, two-tailed nested ANOVA, with Tukey multiple comparison of means. b, Transwell invasion of HCT116 CRC cells after treatment with lactate (positive control), acetate, propionate (10mM), alanine and succinate (5mM), n=three biologically independent experiments, each with three technical replicates, ns=not significant, p=0.0030, p=0.0004, p=0.0217, p=0.0479 for lactate, acetate, propionate and succinate vs. control respectively, ordinary one-way ANOVA Dunnett testing. c, Protein levels of ALDH2 in formate-treated (+) and untreated (-) cells. d, ALDH activity in human colonic organoids after treatment with Fn. Data shows mean±SEM of n=three biologically independent experiments (ns=not significant). e, Wnt activity in 7TGO-RKO Wnt-reporter cells stimulated with bacteria Fn, Gemella morbillorum (Gm) and Parvimonas micra (Pm)) or with 10% [v/v] cell-free bacterial culture supernatants (-sup) or with bacterial secretion products (10mM) for 1h. Bars represent mean±SD of n=one (bacterial supernatants), two (lactate, acetate), three (Gm, Pm, formate, Wnt), four (Fn) or six (control) biological replicates. f, Protein levels of pERK/ERK, pp65/p65 and pp38/p38 in formate treated (+) and untreated (-) cells. g, Transwell invasion of HCT116 CRC cells after 48 hours of stimulation with formate alone or in presence of a Wnt-signaling inhibitor (F535). Data shows mean±SEM of n=three biologically independent experiments. Control and formate condition alone are also shown in Fig. 4b. ns=not significant, repeated measures one-way ANOVA. h, Wnt activity in Fn-supernatant treated 7TGP-RKO Wnt-reporter cells alone or with an AhR inhibitor. Data shows mean±SEM of two biologically independent experiments, p=0.0228. i, Stemness marker CD44 in mouse tumours after intratumoral injection of Fn. Data shows mean±SEM percentage of CD44⁺ live cells (ns=not significant), n=8 independently treated tumors. j, Photograph of explanted subcutaneous xenografts from secondary recipient mice from Fig. 5f. k, Expression of stemness markers in formate-injected and control tumors, n=10 and n=12 independently treated tumors for the control and formate conditions respectively. Data shows mean±SEM for mean fluorescence intensity (MFI) of ALDH and CD44 and percentage of CD24, CD133, Nanog and Sox2, expressing cells. (ns=not significant, p=0.0460 for ALDH); unpaired, two-sided t-test in (d), (i), (j), and (l). *p<0.05, **p<0.01, ***p<0.001. Source data

Extended Data Fig. 6. Fusobacterium nucleatum and formate induced pathways.

a, Representative images of open colons of PBS or Fn treated mice. Black arrows mark tumors, scalebar=5mm. b, Colon length in centimeters (measured upon resection, left) and mouse body weight over time in grams (right) of germ-free mice gavaged with Fn and control mice. Data shows mean±SEM, n=11 and n=10 biologically individual animals for the control and Fn-treated groups respectively, pooled from two independent experiments. c, Representative gating strategy of Th17 cells. Th17 cell gating was performed on live (APC-Cy7^-) single cells with FITC(CD4)⁺, BV605(IL-17)⁺ and APC(RORγT)⁺ staining. d, Gene expression levels of AHR as assessed by RT qPCR in Fn-gavaged mice. Data shows mean±SEM of n=4 biological replicates in control and n=3 biological replicates in Fn, two-sided t-test. Source data

Extended Data Fig. 7. Flow cytometry gating strategies.

a, Representative gating strategy of dendritic cells. b, Representative gating strategy of macrophages. c, Representative gating strategy of T regulatory cells. d, Representative gating strategy of CD8 cells. e, Representative gating strategy of T helper 1 (subpanel 1) and T helper 2 (subpanel 2) cells.

Extended Data Fig. 8. Flow cytometry gating strategies.

a, Representative gating strategy of innate lymphoid cells (ILC); ILC1 (subpanel 1), ILC2 (subpanel 2), ILC3 (subpanel 3). b, Representative gating strategy of natural killer (NK, subpanel 1) and natural killer T (NKT, subpanel 2) cells.